A Method and System for Generating Power Personnel Protection Paths Based on Multi-Objective Optimization

Abstract

This invention relates to the field of path planning technology, and particularly to a method and system for generating protective paths for power personnel based on multi-objective optimization. It constructs an air-to-ground collaborative monitoring platform by combining UAV swarm networking and a multi-source sensing collaborative mechanism; it invokes the air-to-ground collaborative monitoring platform to perform vertical spatial multi-source monitoring of the target substation and performs field strength modeling to obtain a three-dimensional electromagnetic distribution model; it acquires real-time working information of multi-directional optimization targets and target power personnel, and uses a DRL hybrid model to analyze the three-dimensional electromagnetic distribution model to generate a real-time intelligent protection path; based on Kalman filtering and particle swarm optimization algorithms, combined with multi-directional optimization targets, it adjusts the real-time intelligent protection path to obtain an adjusted intelligent protection path. This invention can guide power personnel to avoid areas with electromagnetic hazards, minimize potential safety risks, and improve work efficiency.

Description

Technical Field

[0001] This invention relates to the field of path planning technology, and in particular to a method and system for generating safety paths for power personnel based on multi-objective optimization. Background Technology

[0002] Existing substation electromagnetic monitoring primarily relies on single-point static data acquisition, depending on fixed equipment to collect data from specific areas. Path planning largely depends on manual experience, with some adjustments made based on simple static electromagnetic data. Furthermore, data processing is centralized in the ground control system, resulting in transmission and processing delays and hindering real-time monitoring, analysis, and path adjustment. Current technologies suffer from several problems, including difficulty in balancing path planning safety and efficiency, inability to adapt to dynamic changes in the electromagnetic environment and personnel movement, slow response to potential hazards, and inherent safety risks. Summary of the Invention

[0003] This invention addresses the problems in existing technologies, such as the difficulty in balancing safety and efficiency in path planning, inability to adapt to dynamic changes in the electromagnetic environment and personnel movement, slow response to potential hazards, and the existence of safety risks. It provides a method and system for generating protective paths for power personnel based on multi-objective optimization to solve these problems.

[0004] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0005] In a first aspect, the present invention provides a method for generating protection paths for power personnel based on multi-objective optimization, comprising: constructing an air-ground collaborative monitoring platform by combining UAV swarm networking and multi-source perception collaborative mechanism; calling the air-ground collaborative monitoring platform to perform vertical spatial multi-source monitoring of the target substation and perform field strength modeling to obtain a three-dimensional electromagnetic distribution model; acquiring real-time working information of multi-directional optimization targets and target power personnel, analyzing the three-dimensional electromagnetic distribution model using a DRL hybrid model to generate a real-time intelligent protection path; and adjusting the real-time intelligent protection path based on Kalman filtering and particle swarm optimization algorithms, combined with multi-directional optimization targets, to obtain an adjusted intelligent protection path.

[0006] Optionally, an air-to-ground collaborative monitoring platform can be constructed by combining UAV swarm networking and multi-source sensing collaboration mechanism, including: deploying multiple UAVs and equipping each UAV with multi-source sensing devices to construct a UAV swarm network; using the UAV swarm to collect electromagnetic field strength data, environmental data and equipment status information of the target substation, and transmitting the data to the ground control system for data fusion processing through an anti-interference communication protocol to construct the air-to-ground collaborative monitoring platform.

[0007] Optionally, the multi-source sensing device includes an electromagnetic field sensor, a lidar, and an infrared thermal imager.

[0008] Optionally, multiple drones are deployed, and each drone is equipped with a multi-source sensing device to build a drone cluster network, including: traversing the multiple drones to deploy lightweight edge terminals and obtain multiple edge computing modules; retrieving historical computing resources from the multiple edge computing modules to obtain multiple historical computing resource usage analysis data; and updating the resource configuration of the multiple edge computing modules based on the multiple historical computing resource usage analysis data.

[0009] Optionally, the air-to-ground collaborative monitoring platform is invoked to perform multi-source vertical spatial monitoring of the target substation and to perform field strength modeling to obtain a three-dimensional electromagnetic distribution model. This includes: using a cluster of unmanned aerial vehicles (UAVs) in the air-to-ground collaborative monitoring platform to perform multi-source sensing of the vertical space of the target substation within a preset monitoring window, and transmitting the data after preprocessing by the edge computing module to the ground control system to obtain a set of multi-source sensing data sequences; traversing the set of multi-source sensing data sequences to perform data fusion processing within the sequences to obtain a set of multi-source sensing fused data; and analyzing the set of multi-source sensing fused data through an electromagnetic field modeling algorithm to obtain a three-dimensional electromagnetic distribution network, and performing interpolation analysis on the spatial region of the sampling points to obtain the three-dimensional electromagnetic distribution model.

[0010] The process of traversing the multi-source sensing data sequence set and performing intra-sequence data fusion processing to obtain a multi-source sensing fused data set includes: performing multi-scale feature analysis on the first multi-source sensing data sequence in the multi-source sensing data sequence set to obtain a multi-scale feature set; performing pairwise fine-grained approximation analysis on the multi-scale feature set to construct a pairwise adjacency fusion matrix set; performing convolution on the corresponding two multi-scale features in the multi-scale feature set based on the pairwise adjacency fusion matrix set, and performing mean processing on the convolution results to obtain multi-scale fused features; using the multi-scale fusion features to perform feature inversion enhancement on the first multi-source sensing data at the end of the first multi-source sensing data sequence to obtain the first multi-source sensing fused data, and adding the first multi-source sensing fused data to the multi-source sensing fused data set.

[0011] The process involves acquiring multi-directional optimization targets and real-time work information of target power personnel, analyzing the three-dimensional electromagnetic distribution model using a DRL hybrid model, and generating a real-time intelligent protection path. This includes: acquiring multiple hazard sample libraries to perform reinforcement learning on the DRL-Transformer model to obtain the DRL hybrid model; using the multi-directional optimization targets as constraints, identifying the input three-dimensional electromagnetic distribution model and real-time work information using the DRL hybrid model to obtain the real-time intelligent protection path.

[0012] The method involves adjusting the real-time intelligent protection path based on Kalman filtering and particle swarm optimization algorithms, combined with multi-directional optimization objectives, to obtain an adjusted intelligent protection path. This includes: acquiring the current movement position of the target power personnel; using the Kalman filtering algorithm to smooth the path nodes of the real-time intelligent protection path and the current movement position to obtain an initial adjusted intelligent protection path; acquiring the dynamic change characteristics of the three-dimensional electromagnetic distribution model; using the initial adjusted intelligent protection path as initial particle points, and with the multi-directional optimization objectives as optimization constraints, using the particle swarm optimization algorithm to adjust the initial adjusted intelligent protection path according to the dynamic change characteristics to obtain the adjusted intelligent protection path.

[0013] The multi-directional optimization objectives include minimizing electromagnetic exposure, minimizing the shortest work path, minimizing dwell risk, and improving path connectivity.

[0014] Secondly, the present invention provides a power personnel protection path generation system based on multi-objective optimization, comprising:

[0015] The air-to-ground collaborative monitoring platform construction module is used to build an air-to-ground collaborative monitoring platform by combining UAV swarm networking and multi-source perception collaborative mechanisms.

[0016] The three-dimensional electromagnetic distribution model acquisition module is used to call the air-ground collaborative monitoring platform to perform vertical spatial multi-source monitoring of the target substation, and to perform field strength modeling to obtain a three-dimensional electromagnetic distribution model.

[0017] The real-time intelligent protection path generation module is used to acquire real-time work information of multi-directional optimization targets and target power personnel, and to analyze the three-dimensional electromagnetic distribution model using the DRL hybrid model to generate a real-time intelligent protection path.

[0018] The real-time intelligent protection path adjustment module is used to adjust the real-time intelligent protection path based on Kalman filtering and particle swarm optimization algorithms, combined with multi-directional optimization objectives, to obtain the adjusted intelligent protection path.

[0019] By implementing this invention, an air-ground collaborative monitoring platform can be built by combining UAV swarm networking and multi-source sensing collaboration mechanism, breaking through the limitations of traditional single-point static monitoring, achieving comprehensive coverage of substation monitoring, avoiding the one-sidedness of single-point monitoring, making the monitoring range wider, and being able to capture the situation in different areas of the substation.

[0020] By implementing this invention, it is possible to call the air-ground collaborative monitoring platform to perform multi-source monitoring of the vertical space of the target substation, and to perform field strength modeling to obtain a three-dimensional electromagnetic distribution model. This enables dynamic monitoring of electromagnetic hazards in the vertical space of the substation, filling the gap in the vertical space dimension of traditional monitoring. Traditional monitoring is mostly concentrated at the horizontal level and cannot fully grasp the electromagnetic situation in the vertical space. This step can more accurately reflect the electromagnetic distribution at different heights.

[0021] By implementing this invention, it is possible to obtain real-time work information of multi-directional optimization targets and target power personnel, analyze the three-dimensional electromagnetic distribution model using the DRL hybrid model, generate real-time intelligent protection paths, and comprehensively consider the safety, efficiency and other needs of power personnel in their work. This ensures that the generated protection paths maximize work efficiency while protecting personnel safety, and avoids situations where single-target optimization leads to neglecting one aspect for another.

[0022] By implementing this invention, it is possible to adjust the real-time intelligent protection path based on Kalman filtering and particle swarm optimization algorithm, combined with multi-directional optimization objectives, to obtain an adjusted intelligent protection path. Kalman filtering can effectively handle the uncertainty of path nodes, smoothly correct the path, make the path more in line with the actual movement of personnel, and avoid sudden changes in path nodes that may cause inconvenience to personnel movement or increase safety risks.

[0023] In summary, by implementing this invention, it is possible to guide power personnel to avoid areas with electromagnetic hazards, minimize potential safety risks, and improve work efficiency. Attached Figure Description

[0024] Figure 1 A flowchart illustrating a method for generating protection paths for power personnel based on multi-objective optimization, provided by the present invention;

[0025] Figure 2 This is a schematic diagram of a power personnel protection path generation system based on multi-objective optimization provided by the present invention.

[0026] In the attached diagram, the components represented by each number are as follows:

[0027] The system includes: an air-ground collaborative monitoring platform construction module 11; a three-dimensional electromagnetic distribution model acquisition module 12; a real-time intelligent protection path generation module 13; and a real-time intelligent protection path adjustment module 14. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0030] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0031] Example 1, as Figure 1 As shown, this embodiment of the invention provides a method for generating protection paths for power personnel based on multi-objective optimization, including:

[0032] S100: Construct an air-ground collaborative monitoring platform by combining UAV swarm networking and multi-source sensing collaboration mechanisms;

[0033] S200: Call the air-ground collaborative monitoring platform to perform vertical spatial multi-source monitoring of the target substation, and perform field strength modeling to obtain a three-dimensional electromagnetic distribution model;

[0034] S300: Acquire real-time work information of multi-directional optimization targets and target power personnel, analyze the three-dimensional electromagnetic distribution model using the DRL hybrid model, and generate real-time intelligent protection paths;

[0035] S400: Based on Kalman filtering and particle swarm optimization algorithms, it adjusts the real-time intelligent protection path by combining multi-directional optimization objectives to obtain an adjusted intelligent protection path.

[0036] In step S100 of this application embodiment, an air-to-ground collaborative monitoring platform is constructed by combining UAV swarm networking and multi-source perception collaboration mechanisms, including:

[0037] Deploy multiple drones and equip each drone with multi-source sensing devices to build a drone cluster network;

[0038] The drone swarm collects electromagnetic field strength data, environmental data, and equipment status information of the target substation and transmits them to the ground control system for data fusion processing via an anti-interference communication protocol to construct the air-ground collaborative monitoring platform.

[0039] The multi-source sensing devices include electromagnetic field sensors, lidar, and infrared thermal imagers.

[0040] In this embodiment of the application, the objective of step S100 is to provide high-precision, real-time, and multi-dimensional data acquisition and transmission support for subsequent monitoring of electromagnetic hazards in the vertical space of substations, three-dimensional electromagnetic distribution modeling, and generation of personnel protection paths.

[0041] To achieve the above objectives, firstly, multiple drones need to be deployed, and each drone needs to be equipped with multi-source sensing devices to build a drone swarm network.

[0042] In step S100 of this application embodiment, multiple drones are deployed, and each drone is equipped with a multi-source sensing device to construct a drone cluster network, including:

[0043] By traversing the multiple drones to deploy lightweight edge terminals, multiple edge computing modules are obtained;

[0044] Historical computing resource retrieval is performed on the multiple edge computing modules to obtain historical computing resource usage analysis data.

[0045] The resource configuration of the multiple edge computing modules is updated based on the analysis data of the multiple historical computing resource usage.

[0046] First, it is necessary to deploy lightweight edge terminals across the multiple drones to obtain multiple edge computing modules. Specifically, based on the drone's load capacity and monitoring requirements, lightweight edge terminals with "low power consumption and high computing density" are selected. For example, a lightweight edge terminal with "power consumption <15W" can be selected. This terminal needs to support multiple interface expansion and can be directly connected to multi-source sensing devices such as electromagnetic field sensors, lidar, and infrared thermal imagers.

[0047] Next, each drone is traversed, and the edge terminal is integrated into the drone body through mounting or embedded installation. Communication debugging between the terminal and the sensing device is completed, such as through RS485 or Ethernet protocol to achieve real-time data interaction, ensuring that each drone has an independent computing power unit for data acquisition and local preprocessing, thus forming an edge computing module.

[0048] Then, it is necessary to perform historical computing resource retrieval on the multiple edge computing modules to obtain multiple historical computing resource usage analysis data.

[0049] The historical computing resource retrieval targets the historical operating data of all deployed edge computing modules in past substation monitoring tasks. If it is the first deployment, the historical monitoring data of the same type of substation and the same type of UAV can be referenced.

[0050] Historical computing resource usage analysis data mainly includes computing load, processing efficiency, and resource idleness. Computing load can be measured by CPU / GPU utilization and memory usage under different monitoring scenarios; processing efficiency can be measured by the preprocessing time of a single multi-source data point and the amount of data processed per unit time; resource idleness can be measured by the proportion of idle computing power and storage resource utilization under low-load scenarios.

[0051] Furthermore, the resource configuration of the multiple edge computing modules needs to be updated based on the analysis data of the multiple historical computing resource usage.

[0052] Specifically, resources are allocated based on the principle of matching demand.

[0053] For high-demand edge computing modules, such as those whose historical computing resource usage analysis data shows that the computing load is consistently >80%, these modules are mostly located in areas with dense equipment and complex electromagnetic environments. By using the "dynamic resource scheduling function" of the edge terminal, the number of CPU cores can be increased, the memory capacity can be expanded, or the data processing algorithm can be optimized to improve processing efficiency.

[0054] Low-demand edge computing modules, such as those whose historical computing resource usage analysis data shows that the computing load is consistently less than 30%, are mostly located in open areas. To reduce unnecessary computing power allocation, such as shutting down some idle CPU cores, the terminal power consumption can be reduced and the drone's flight time extended.

[0055] Meanwhile, for collaborative monitoring tasks of drone swarms, ensure that the computing power configuration of adjacent edge computing modules is balanced to avoid "data synchronization delay" caused by the slow processing of a certain module. For example, ensure that the computing power of edge computing modules at heights of 30 meters and 35 meters is consistent to ensure that data is transmitted to the ground at the same time.

[0056] Furthermore, it is necessary to utilize the aforementioned drone swarm to collect electromagnetic field strength data, environmental data, and equipment status information of the target substation.

[0057] Specifically, the focus can be on the vertical space of the target substation from 5 to 40 meters, and monitoring nodes can be divided according to the vertical gradient, such as setting a sampling point every 1 meter to avoid missed measurements.

[0058] The types of data collected include electromagnetic field strength data, environmental data, and equipment status information of the target substation. Electromagnetic field strength data is collected in real time by electromagnetic field sensors to capture the radiation intensity at each monitoring point, which is used for subsequent construction of a three-dimensional electromagnetic distribution model. Environmental data specifically includes temperature, humidity, and air pressure, used to correct for interference from environmental factors affecting the electromagnetic data.

[0059] The equipment status information is obtained by using an infrared thermal imager to capture potential hazards such as abnormal temperature and damaged insulation layers in equipment such as transformers and switchgear, and simultaneously record equipment operating parameters such as voltage and current.

[0060] Finally, the data needs to be transmitted to the ground control system via an anti-interference communication protocol for data fusion processing to construct the air-ground collaborative monitoring platform.

[0061] If the transmission distance is long or there is signal obstruction, the signal can be enhanced by "relay drones" in the drone swarm to ensure transmission stability.

[0062] The data transmitted to the ground undergoes "multi-source heterogeneous data fusion". Specifically, the raw data is first preprocessed by the edge computing module to remove outliers and standardize the format. Then, electromagnetic field strength data, environmental data, and equipment status data are fused to eliminate data redundancy and conflicts. The fused data is stored in the ground control system, forming a closed loop with the real-time data acquisition of the UAV swarm, and finally constructing an air-ground collaborative monitoring platform.

[0063] In step S200 of this application embodiment, the air-ground collaborative monitoring platform is invoked to perform vertical spatial multi-source monitoring of the target substation and to perform field strength modeling to obtain a three-dimensional electromagnetic distribution model, including:

[0064] The UAV cluster network in the air-ground collaborative monitoring platform is used to perform multi-source sensing of the vertical space of the target substation within a preset monitoring window, and the data after preprocessing by the edge computing module is transmitted to the ground control system to obtain a set of multi-source sensing data sequences.

[0065] The multi-source sensing data sequence set is traversed to perform intra-sequence data fusion processing to obtain a multi-source sensing fused data set.

[0066] The multi-source sensing fusion data set is analyzed by electromagnetic field modeling algorithm to obtain a three-dimensional electromagnetic distribution network, and the spatial region of the sampling points is interpolated to obtain the three-dimensional electromagnetic distribution model.

[0067] In this embodiment of the application, the purpose of step S200 is to transform the multi-source raw data collected by the air-ground collaborative monitoring platform into a visualized, high-precision, and fully covered three-dimensional electromagnetic distribution model, so as to provide a spatial electromagnetic risk map for the subsequent generation of personnel protection paths based on the DRL hybrid model.

[0068] To achieve the above objectives, it is first necessary to utilize the UAV cluster network in the air-ground collaborative monitoring platform to perform multi-source sensing of the vertical space of the target substation within a preset monitoring window, and transmit the data after preprocessing by the edge computing module to the ground control system to obtain a set of multi-source sensing data sequences.

[0069] The preset monitoring window can be set as a dual-dimensional monitoring window of "time + space" according to the vertical spatial range and monitoring accuracy requirements of the substation. For example, "collect data once every 10 seconds, set a sampling point every 0.5 meters in the vertical direction, and set a sampling point every 2 meters in the horizontal direction according to the equipment layout", to ensure coverage without blind spots.

[0070] Next, the drone swarm needs to operate synchronously according to the monitoring window. Each drone uses multi-source sensing devices such as electromagnetic field sensors, lidar, and infrared thermal imagers to collect parameters such as electromagnetic field strength, spatial coordinates, equipment temperature, and ambient humidity / temperature at its corresponding sampling point, forming a preliminary set of single-sampling-point, multi-type raw data. After preprocessing operations such as outlier removal, data standardization, and timestamp synchronization by the edge computing module, the data is transmitted to the ground control system. The data is then sorted by sampling time and spatial coordinates to form a multi-source sensing data sequence set. For example, "Time T1 - Coordinates (X1, Y1, 30 meters) - Electromagnetic strength 5.2V / m - Equipment temperature 38℃ - Humidity 60%" constitutes a data sequence. All sequences are then aggregated to form the multi-source sensing data sequence set.

[0071] Furthermore, it is necessary to traverse the multi-source sensing data sequence set to perform intra-sequence data fusion processing to obtain a multi-source sensing fusion data set.

[0072] In step S200 of this application embodiment, the multi-source sensing data sequence set is traversed to perform intra-sequence data fusion processing to obtain a multi-source sensing fusion data set, including:

[0073] Multi-scale feature analysis is performed on the first multi-source sensing data sequence in the multi-source sensing data sequence set to obtain a multi-scale feature set.

[0074] Perform pairwise fine-grained approximation analysis on the multi-scale feature set to construct a pairwise adjacency fusion matrix set;

[0075] Based on the pairwise adjacency fusion matrix set, convolution is performed on the corresponding two multi-scale features in the multi-scale feature set, and the convolution results are averaged to obtain multi-scale fusion features.

[0076] The first multi-source sensing data located at the end of the first multi-source sensing data sequence is enhanced by using the multi-scale fusion features to obtain the first multi-source sensing fusion data, and the first multi-source sensing fusion data is added to the multi-source sensing fusion data set.

[0077] First, multi-scale feature analysis needs to be performed on the first multi-source sensing data sequence in the multi-source sensing data sequence set to obtain a multi-scale feature set.

[0078] Specifically, it is necessary to extract features from multiple dimensions according to data type and scale, such as electromagnetic field strength data, equipment temperature data, and environmental humidity data.

[0079] The electromagnetic field strength data is extracted from multiple scales, including instantaneous, short-term, and long-term features. These include instantaneous peak values ​​(e.g., the maximum electromagnetic field strength within 10ms), short-term average values ​​(e.g., the average electromagnetic field strength within 1 minute), and trend features (e.g., the rate of change compared to the previous 5 minutes). Equipment temperature data requires the extraction of single-point, area, and change features, such as real-time temperature at a single point, the average temperature of three sampling points around the equipment, and the rate of temperature increase (degrees per minute). Environmental humidity data requires the extraction of deviation features between the current value and a threshold, such as the real-time humidity value and the deviation from the substation's safe humidity threshold. For example, the substation's safe humidity threshold could be 70%.

[0080] All extracted features are categorized and summarized according to "data type - scale" to form a multi-scale feature set corresponding to the "first multi-source sensing data sequence", such as "instantaneous peak electromagnetic intensity 5.8V / m, short-term average electromagnetic intensity 5.2V / m, equipment temperature 39℃, humidity deviation -5%".

[0081] Furthermore, it is necessary to perform pairwise fine-grained approximation analysis on the multi-scale feature set to construct a pairwise adjacency fusion matrix set.

[0082] First, select "highly correlated feature pairs": select features with physical correlation from the multi-scale feature set and combine them in pairs, such as "short-term mean of electromagnetic intensity and equipment temperature" and "instantaneous peak value of electromagnetic intensity and humidity deviation", and exclude uncorrelated feature pairs;

[0083] Then, using either cosine similarity or Pearson correlation coefficient, the correlation degree of each pair of features is calculated. For example, through statistical analysis of historical monitoring data, it was found that the correlation coefficient between "short-term average electromagnetic intensity and equipment temperature" is 0.85 (strong correlation), and the correlation coefficient between "instantaneous peak electromagnetic intensity and humidity deviation" is 0.3 (weak correlation). The correlation degree value is directly used as the weight ratio when fusing the pair of features; the stronger the correlation, the higher the weight, ensuring that the fusion prioritizes the reflection of key correlations.

[0084] Next, a matrix is ​​constructed with the "feature pair-correlation degree" relationship as the core: for each pair of features, the correlation degree is used as the matrix element value to form a "pairwise adjacent fusion matrix". For example, the matrix element of "short-term mean electromagnetic intensity (feature A) and equipment temperature (feature B)" is 0.85, and the matrix element of "instantaneous peak electromagnetic intensity (feature C) and humidity deviation (feature D)" is 0.3.

[0085] Finally, the fusion matrices of all feature pairs are summarized to form a set of pairwise adjacent fusion matrices, clarifying the weight ratio of each feature pair during fusion.

[0086] Furthermore, it is necessary to perform convolution on the two corresponding multi-scale features in the multi-scale feature set based on the pairwise adjacency fusion matrix set, and then perform mean processing on the convolution results to obtain multi-scale fusion features.

[0087] First, feature pairs are matched and convolved according to the "fusion matrix weights": For each matrix in the set of pairwise adjacent fusion matrices, two corresponding multi-scale features are extracted, such as feature A and feature B. The correlation in the fusion matrix is ​​used as the "convolution kernel weights". Convolution operation is performed on the values ​​of the two features. For example, if feature A = 5.2V / m and feature B = 39℃, the convolution calculation is 5.2×0.85 + 39×0.85, which strengthens the synergistic effect of strongly correlated features.

[0088] Repeat the above convolution operation for all highly correlated feature pairs to obtain multiple convolution results.

[0089] Take the arithmetic mean of all convolution results. For example, if the three convolution results are 30.94, 8.6, and 12.3, the mean is (30.94+8.6+12.3) / 3≈17.28. This mean is the "multi-scale fusion feature". Its essence is to comprehensively reflect the overall state of the electromagnetic-equipment-environment at the sampling point by balancing the weights and the mean, and avoid the excessive dominance of a single feature or strongly correlated feature.

[0090] Furthermore, the multi-scale fusion features are used to perform feature inversion enhancement on the first multi-source sensing data located at the end of the first multi-source sensing data sequence to obtain the first multi-source sensing fusion data, and the first multi-source sensing fusion data is added to the multi-source sensing fusion data set.

[0091] Specifically, a "fusion feature-original data" inversion model needs to be established. This involves training the inversion model using linear regression or a lightweight neural network based on the correspondence between historical fusion data and corrected original data, thus clarifying the mapping relationship between multi-scale fusion features and the correction amount of the original data. Then, the multi-scale fusion features obtained in the preceding steps are input into the inversion model to calculate the correction amount of the original data. For example, if the original electromagnetic field strength data is 5.1 V / m, the inversion model outputs a correction amount of +0.1 V / m, resulting in a corrected data of 5.2 V / m.

[0092] The corrected core data, such as 5.2V / m, is integrated with other key information from the sequence, such as sampling coordinates, timestamps, and equipment temperature of 39℃, to form the "first multi-source sensing fusion data". This first multi-source sensing fusion data is added to the multi-source sensing fusion data set. The above steps are repeated to traverse all multi-source sensing data sequences, ultimately forming a multi-source sensing fusion data set covering all sampling points in the substation.

[0093] Furthermore, it is necessary to analyze the multi-source sensing fusion data set through electromagnetic field modeling algorithms to obtain a three-dimensional electromagnetic distribution network, and to perform interpolation analysis on the spatial region of the sampling points to obtain the three-dimensional electromagnetic distribution model.

[0094] Specifically, the "electromagnetic intensity data and spatial coordinate data" from the multi-source sensing fusion dataset need to be input into the "electromagnetic field modeling algorithm." The electromagnetic field modeling algorithm, such as the finite element method or boundary element method, is an existing technology and will not be elaborated upon here. The electromagnetic field modeling algorithm calculates the electromagnetic radiation propagation patterns under different spatial coordinates, and initially constructs a "three-dimensional electromagnetic distribution network"—this network contains the correspondence between the electromagnetic intensity values ​​and spatial locations of all sampling points, but there are "data gaps" between sampling points.

[0095] Then, for the blank areas in the three-dimensional electromagnetic distribution network, such as the space between two sampling points, a "spatial interpolation algorithm", such as Kriging interpolation, is used to ensure that the interpolation accuracy deviates from the actual value by <0.1V / m. Based on the electromagnetic intensity data of the surrounding sampling points, the electromagnetic intensity value of the blank area is calculated to achieve "full-domain spatial electromagnetic data completion".

[0096] Finally, the completed global electromagnetic data is combined with the substation's three-dimensional spatial model, including equipment layout and building structure models, and visualization techniques such as heatmap rendering are used to generate a "three-dimensional electromagnetic distribution model." This model can visually display the electromagnetic radiation intensity at any location within a 5-40 meter vertical space of the substation. For example, red areas indicate high risk, corresponding to an electric field strength >8V / m; green areas indicate safety, corresponding to an electric field strength <3V / m. This provides a direct basis for subsequent analysis of operational risks and the generation of protective paths.

[0097] In step S300 of this application embodiment, real-time work information of multi-directional optimization targets and target power personnel is obtained, and the three-dimensional electromagnetic distribution model is analyzed using a DRL hybrid model to generate a real-time intelligent protection path, including:

[0098] Multiple hazard sample databases are obtained to perform reinforcement learning on the DRL-Transformer model, thereby obtaining the DRL hybrid model;

[0099] Using the multi-directional optimization objective as a constraint, the DRL hybrid model is used to identify the input three-dimensional electromagnetic distribution model and real-time working information to obtain the real-time intelligent protection path.

[0100] In this embodiment of the application, the purpose of step S300 is to combine the spatial risk information of the three-dimensional electromagnetic distribution model with the real-time working status of personnel through the intelligent learning and multi-objective constraint optimization of the DRL hybrid model, and output a protection path that takes into account "safety, efficiency and practicality", so as to provide accurate guidance for power personnel to avoid electromagnetic hazards and reduce operational risks.

[0101] First, multiple hazard sample databases need to be acquired to perform reinforcement learning on the DRL-Transformer model, resulting in the DRL hybrid model. These databases contain data on electromagnetic distribution, equipment hazards, and historical optimal paths for different substations. The DRL-Transformer model is trained using these samples: the Transformer layer first associates features from the electromagnetic and path data, while the DRL layer learns path decisions based on objectives such as "less radiation and shorter path." Parameters are repeatedly adjusted until the model can accurately assess risks and output the optimal path; at this point, the model is the DRL hybrid model.

[0102] Next, constrained by the multi-directional optimization objective, the DRL hybrid model is used to identify the input three-dimensional electromagnetic distribution model and real-time working information to obtain the real-time intelligent protection path.

[0103] Specifically, key data needs to be input into the DRL hybrid model, namely the three-dimensional electromagnetic distribution model of the substation, the current location of personnel, and real-time work information such as the work point obtained in step S200.

[0104] The DRL hybrid model is constrained by "multi-directional optimization objectives". It first associates electromagnetic risks and personnel location characteristics through the Transformer layer, then uses the DRL layer to calculate the path "from the current location to the work point", ensuring that the path avoids high-risk areas and meets safety and efficiency requirements. Finally, it outputs this real-time intelligent protection path.

[0105] In step S400 of this application embodiment, the real-time intelligent protection path is adjusted based on Kalman filtering and particle swarm optimization algorithms, combined with multi-directional optimization objectives, to obtain the adjusted intelligent protection path, including:

[0106] The current movement position of the target power personnel is obtained, and the Kalman filter algorithm is used to smooth the path nodes of the real-time intelligent protection path and the current movement position to obtain the initial adjusted intelligent protection path.

[0107] Obtain the dynamic variation characteristics of the three-dimensional electromagnetic distribution model;

[0108] Using the initial adjusted intelligent protection path as the initial particle point, and with the multi-directional optimization objective as the optimization constraint, the particle swarm optimization algorithm is used to adjust the initial adjusted intelligent protection path according to the dynamic change characteristics to obtain the adjusted intelligent protection path.

[0109] In step S400 of this application embodiment, the multi-directional optimization objectives include minimizing electromagnetic exposure, minimizing the shortest operation path, minimizing dwell risk, and path connectivity.

[0110] In this embodiment of the application, the purpose of step S400 is to solve the problem of adapting the "staticity" of the path to the "dynamic changes" of the site. Through the smoothing correction of Kalman filtering and the dynamic adjustment of particle swarm optimization, the final output is an intelligent protection path that fits the real-time position of the personnel, responds to changes in the electromagnetic environment, and meets multiple safety and efficiency objectives.

[0111] To achieve the above objectives, it is first necessary to obtain the current movement location of the target power personnel, and then use the Kalman filter algorithm to perform path node smoothing correction on the real-time intelligent protection path and the current movement location to obtain the initial adjusted intelligent protection path.

[0112] Among them, the current movement position of personnel can be collected in real time by the positioning device carried by the personnel, such as GPS, UWB positioning, etc., and the current coordinates of the target power personnel can be compared with the preset nodes of the original real-time intelligent protection path to identify position deviations.

[0113] Then, the Kalman filter algorithm is used to process the preset node and current position data of the personnel in the original real-time intelligent protection path. That is, the theoretical position of the personnel in the next moment is predicted first, and then the prediction deviation is corrected by combining the actual collected position data to eliminate positioning noise. Based on the corrected position, the node order and coordinates of the original path are adjusted. For example, the node "X4,Y6→X6,Y8" in the original path is corrected to "X5,Y7→X6,Y8" to ensure that the path and the actual movement trajectory of the personnel are smoothly connected, and the initial adjusted intelligent protection path is obtained.

[0114] Furthermore, it is necessary to obtain the dynamic change characteristics of the three-dimensional electromagnetic distribution model. Specifically, this can be achieved by using a cluster of drones on an air-to-ground collaborative monitoring platform to collect data in real time on changes in electromagnetic intensity, equipment status, equipment data, and environmental interference in the changing areas, thereby updating the three-dimensional electromagnetic distribution model.

[0115] Furthermore, the initial intelligent protection path needs to be used as the initial particle point, with multi-directional optimization objectives as optimization constraints. The particle swarm optimization algorithm is then used to adjust the initial intelligent protection path according to the dynamic change characteristics to obtain the adjusted intelligent protection path.

[0116] Specifically, the "initial adjustment of intelligent protection path" obtained from the aforementioned steps needs to be used as the initial particle of the particle swarm algorithm, i.e., the initial solution;

[0117] Then set optimization objectives: Objective 1: Minimize the total electromagnetic exposure along the entire path, i.e., minimize electromagnetic exposure; Objective 2: Minimize the total path length, i.e., the shortest operating path; Objective 3: Minimize the dwell time in high-risk areas, such as electromagnetic intensity > 8V / m or equipment overheating areas, i.e., minimize dwell risk; Objective 4: Ensure connectivity between all nodes on the path, i.e., path connectivity.

[0118] Then, the particle swarm optimization algorithm treats each "possible path scheme" as a "particle" and calculates the "fitness" of each particle based on the dynamic change characteristics of the three-dimensional electromagnetic distribution model, which is the degree to which multiple optimization objectives are satisfied. The higher the fitness, the better.

[0119] Each particle references its own historical best path and the group's best path to adjust its own path nodes, such as avoiding high-risk areas and shortening detour distances.

[0120] When the particle swarm finds the path with the "highest fitness," that is, one that satisfies all optimization objectives and has no significant room for improvement, it stops iterating. This path is the final adjusted intelligent protection path.

[0121] Example 2, as Figure 2 As shown, based on the same inventive concept as the power personnel protection path generation method based on multi-objective optimization provided in Embodiment 1, this embodiment of the invention also provides a power personnel protection path generation system based on multi-objective optimization, including:

[0122] Air-ground collaborative monitoring platform construction module 11 is used to construct an air-ground collaborative monitoring platform by combining UAV swarm networking and multi-source perception collaborative mechanism;

[0123] The three-dimensional electromagnetic distribution model acquisition module 12 is used to call the air-ground collaborative monitoring platform to perform vertical spatial multi-source monitoring of the target substation, and to perform field strength modeling to obtain a three-dimensional electromagnetic distribution model.

[0124] The real-time intelligent protection path generation module 13 is used to acquire real-time work information of multi-directional optimization targets and target power personnel, and to analyze the three-dimensional electromagnetic distribution model using the DRL hybrid model to generate a real-time intelligent protection path.

[0125] The real-time intelligent protection path adjustment module 14 is used to adjust the real-time intelligent protection path based on Kalman filtering and particle swarm optimization algorithms, combined with multi-directional optimization objectives, to obtain the adjusted intelligent protection path.

[0126] Furthermore, the air-to-ground collaborative monitoring platform construction module 11 includes the following execution steps:

[0127] Deploy multiple drones and equip each drone with multi-source sensing devices to build a drone cluster network;

[0128] The drone swarm collects electromagnetic field strength data, environmental data, and equipment status information of the target substation, and transmits the data to the ground control system for data fusion processing through an anti-interference communication protocol to construct the air-ground collaborative monitoring platform.

[0129] This involves deploying multiple drones, equipping each drone with multi-source sensing devices, and constructing a drone swarm network, including:

[0130] By traversing the multiple drones to deploy lightweight edge terminals, multiple edge computing modules are obtained;

[0131] Historical computing resource retrieval is performed on the multiple edge computing modules to obtain historical computing resource usage analysis data.

[0132] The resource configuration of the multiple edge computing modules is updated based on the analysis data of the multiple historical computing resource usage.

[0133] Furthermore, the three-dimensional electromagnetic distribution model acquisition module 12 includes the following execution steps:

[0134] The UAV cluster network in the air-ground collaborative monitoring platform is used to perform multi-source sensing of the vertical space of the target substation within a preset monitoring window, and the data after preprocessing by the edge computing module is transmitted to the ground control system to obtain a set of multi-source sensing data sequences.

[0135] The multi-source sensing data sequence set is traversed to perform intra-sequence data fusion processing to obtain a multi-source sensing fused data set.

[0136] The multi-source sensing fusion data set is analyzed by electromagnetic field modeling algorithm to obtain a three-dimensional electromagnetic distribution network, and the spatial region of the sampling points is interpolated to obtain the three-dimensional electromagnetic distribution model.

[0137] Specifically, the process involves traversing the multi-source sensing data sequence set and performing intra-sequence data fusion processing to obtain a multi-source sensing fusion data set, including:

[0138] Multi-scale feature analysis is performed on the first multi-source sensing data sequence in the multi-source sensing data sequence set to obtain a multi-scale feature set.

[0139] Perform pairwise fine-grained approximation analysis on the multi-scale feature set to construct a pairwise adjacency fusion matrix set;

[0140] Based on the pairwise adjacency fusion matrix set, convolution is performed on the corresponding two multi-scale features in the multi-scale feature set, and the convolution results are averaged to obtain multi-scale fusion features.

[0141] The first multi-source sensing data located at the end of the first multi-source sensing data sequence is enhanced by using the multi-scale fusion features to obtain the first multi-source sensing fusion data, and the first multi-source sensing fusion data is added to the multi-source sensing fusion data set.

[0142] Furthermore, the real-time intelligent protection path generation module 13 includes the following execution steps:

[0143] Multiple hazard sample databases are obtained to perform reinforcement learning on the DRL-Transformer model, thereby obtaining the DRL hybrid model;

[0144] Using the multi-directional optimization objective as a constraint, the DRL hybrid model is used to identify the input three-dimensional electromagnetic distribution model and real-time working information to obtain the real-time intelligent protection path.

[0145] Furthermore, the real-time intelligent protection path adjustment module 14 includes the following execution steps:

[0146] The current movement position of the target power personnel is obtained, and the Kalman filter algorithm is used to smooth the path nodes of the real-time intelligent protection path and the current movement position to obtain the initial adjusted intelligent protection path.

[0147] Obtain the dynamic variation characteristics of the three-dimensional electromagnetic distribution model;

[0148] Using the initial adjusted intelligent protection path as the initial particle point, and with the multi-directional optimization objective as the optimization constraint, the particle swarm optimization algorithm is used to adjust the initial adjusted intelligent protection path according to the dynamic change characteristics to obtain the adjusted intelligent protection path.

[0149] The multi-directional optimization objectives include minimizing electromagnetic exposure, minimizing the shortest work path, minimizing dwell risk, and improving path connectivity.

[0150] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0151] Those skilled in the art will understand that embodiments of the present invention can provide methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0152] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0153] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0154] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0155] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Clearly, those skilled in the art can make various alterations and variations to the invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the invention and its equivalents, the invention is also intended to include these modifications and variations.

Claims

1. A method for generating a protection path for a power worker based on multi-objective optimization, characterized in that, The method comprises: An air-ground cooperative monitoring platform is constructed by combining UAV cluster networking and multi-source perception coordination mechanism; A vertical space multi-source monitoring of the target substation is performed by calling the air-ground cooperative monitoring platform, field strength modeling is performed, and a three-dimensional electromagnetic distribution model is obtained; Real-time working information of the multi-direction optimization target and the target electric power personnel is obtained, the three-dimensional electromagnetic distribution model is analyzed by using a DRL hybrid model, and a real-time intelligent protection path is generated; Based on Kalman filtering and particle swarm optimization algorithm, the real-time intelligent protection path is adjusted based on the multi-direction optimization target, and an adjusted intelligent protection path is obtained; The vertical space multi-source monitoring of the target substation is performed by calling the air-ground cooperative monitoring platform, the field strength modeling is performed, and the three-dimensional electromagnetic distribution model is obtained, which comprises: Multi-source perception is performed on the vertical space of the target substation within a preset monitoring window by using the UAV cluster networking in the air-ground cooperative monitoring platform, and the data preprocessed by the edge computing module is transmitted to the ground control system to obtain a multi-source perception data sequence set; In-sequence data fusion processing is performed on the multi-source perception data sequence set to obtain a multi-source perception fusion data set; The multi-source perception fusion data set is analyzed by using an electromagnetic field modeling algorithm, a three-dimensional electromagnetic distribution network is obtained, and interpolation analysis is performed on the sampling point space region to obtain the three-dimensional electromagnetic distribution model; The in-sequence data fusion processing is performed on the multi-source perception data sequence set to obtain the multi-source perception fusion data set, which comprises: Multi-scale feature analysis is performed on a first multi-source perception data sequence in the multi-source perception data sequence set to obtain a multi-scale feature set; Two-by-two fine-grained approximate analysis is performed on the multi-scale feature set to construct a two-by-two adjacent fusion matrix set; Convolution is performed on corresponding two multi-scale features in the multi-scale feature set based on the two-by-two adjacent fusion matrix set, and mean value processing is performed on the convolution result to obtain multi-scale fusion features; Feature inversion enhancement is performed on a first multi-source perception data located at the end in the first multi-source perception data sequence by using the multi-scale fusion features to obtain a first multi-source perception fusion data, and the first multi-source perception fusion data is added to the multi-source perception fusion data set.

2. The multi-objective optimization based power worker protection path generation method of claim 1, wherein, The air-ground cooperative monitoring platform is constructed by combining UAV cluster networking and multi-source perception coordination mechanism, which comprises: A plurality of UAVs are deployed, and each UAV is equipped with a multi-source perception device to construct a UAV cluster network; Electromagnetic field strength data, environmental data and device state information of the target substation are collected by using the UAV cluster, and are transmitted to the ground control system for data fusion processing through an anti-interference communication protocol to construct the air-ground cooperative monitoring platform.

3. The multi-objective optimization based power worker protection path generation method of claim 2, wherein, The multi-source perception device comprises an electromagnetic field sensor, a laser radar and an infrared thermal imager.

4. The multi-objective optimization based power worker protection path generation method of claim 2, wherein, The plurality of UAVs are deployed, and each UAV is equipped with a multi-source perception device to construct a UAV cluster network, which comprises: Lightweight edge terminal deployment is performed on the plurality of UAVs to obtain a plurality of edge computing modules; The multiple edge computing modules are subjected to historical computing resource retrieval to obtain multiple historical computing resource usage analysis data; The multiple edge computing modules are subjected to resource update configuration according to the multiple historical computing resource usage analysis data.

5. The multi-objective optimization based power worker protection path generation method of claim 1, wherein, Real-time working information of target power personnel is obtained, and the three-dimensional electromagnetic distribution model is analyzed by using a DRL hybrid model to generate a real-time intelligent protection path, including: A DRL-Transformer model is subjected to enhanced learning by using multiple hidden danger sample libraries to obtain the DRL hybrid model; The DRL hybrid model is used to identify the input three-dimensional electromagnetic distribution model and real-time working information by taking the multiple direction optimization targets as constraints to obtain the real-time intelligent protection path.

6. The multi-objective optimization based power worker protection path generation method of claim 1, wherein, Based on Kalman filtering and particle swarm optimization algorithm, the real-time intelligent protection path is adjusted by combining the multiple direction optimization targets to obtain an adjusted intelligent protection path, including: The current moving position of the target power personnel is obtained, and the real-time intelligent protection path and the current moving position are subjected to path node smoothing correction by using the Kalman filtering algorithm to obtain an initial adjusted intelligent protection path; Dynamic change characteristics of the three-dimensional electromagnetic distribution model are obtained; The initial adjusted intelligent protection path is taken as an initial particle point, and the particle swarm optimization algorithm is used to adjust the initial adjusted intelligent protection path according to the dynamic change characteristics by taking the multiple direction optimization targets as optimization constraints to obtain the adjusted intelligent protection path.

7. The multi-objective optimization based power worker protection path generation method of claim 1, wherein, The multiple direction optimization targets include minimizing electromagnetic exposure, shortest work path, minimizing staying risk, and path connectivity.

8. A power worker protection path generation system based on multi-objective optimization, characterized by, The system is used to implement the power personnel protection path generation method based on multi-objective optimization, and includes: An air-ground collaborative monitoring platform construction module is used to construct an air-ground collaborative monitoring platform by combining unmanned aerial vehicle cluster networking and multi-source perception collaborative mechanism; A three-dimensional electromagnetic distribution model acquisition module is used to call the air-ground collaborative monitoring platform to perform vertical space multi-source monitoring on a target substation and perform field strength modeling to obtain a three-dimensional electromagnetic distribution model; A real-time intelligent protection path generation module is used to obtain multiple direction optimization targets and real-time working information of target power personnel, and analyze the three-dimensional electromagnetic distribution model by using a DRL hybrid model to generate a real-time intelligent protection path; A real-time intelligent protection path adjustment module is used to adjust the real-time intelligent protection path by combining the multiple direction optimization targets based on Kalman filtering and particle swarm optimization algorithm to obtain an adjusted intelligent protection path.

Images